A dating app for your GPU and local LLMs

What it does

LLM Checker is a Node.js CLI that probes your machine for CPU, RAM, GPU type (NVIDIA, AMD, Intel Arc, Apple Silicon), and acceleration backend. It cross-references that against a packaged SQLite catalog of 229 Ollama models and 7,176 variants, then scores compatibility across four axes: Quality, Speed, Fit, and Context. The output is a ranked list of what you can actually run locally, plus optional ai-run execution with live tokens-per-second metrics.

The interesting bit

The calibration layer. You can run a prompt suite through candidate models, generate a policy file, and then have future recommend and ai-run calls route through that empirical data instead of static heuristics. It also exposes an MCP server, so Claude Code can query your hardware, pull models, and benchmark them without leaving the chat.

Key highlights

• Ships with a prebuilt SQLite catalog (229 models, 7,176 variants) and a sync command to refresh from Ollama’s live registry
• Hardware detection covers Apple Silicon, NVIDIA CUDA, AMD ROCm, Intel Arc, plus integrated vs. dedicated GPU visibility
• Memory estimation uses a bytes-per-parameter formula validated against real Ollama model sizes
• Pure JavaScript, zero native dependencies beyond optional sql.js
• MCP server exposes 15+ tools for hardware analysis, model management, benchmarking, and head-to-head comparison

Caveats

• The ollama_plan_env tool returns “ready-to-paste” shell exports, which feels like a leaky abstraction for a CLI tool
• --mode full currently requires --runtime ollama, so calibration flexibility is narrower than the marketing suggests
• The README warns that the GitHub Packages mirror lags npm, which implies release hygiene is still being figured out

Verdict

Worth a look if you maintain multiple machines with different GPUs and want a single command to answer “will this model run here?” Skip it if you already have a fixed workflow and don’t need another abstraction layer over ollama list.